Department of Quantitative Social Science
Disadvantaged children’s “low”
educational expectations: Are the US and
UK really so different to other
industrialized nations?
John Jerrim
DoQSS Working Paper No. 11-04
June 2011
DISCLAIMER
Any opinions expressed here are those of the author(s) and not
those of the Institute of Education. Research published in this
series may include views on policy, but the institute itself takes no
institutional policy positions.
DoQSS Workings Papers often represent preliminary work and
are circulated to encourage discussion. Citation of such a paper
should account for its provisional character. A revised version may
be available directly from the author.
Department of Quantitative Social Science.
Institute of
Education, University of London. 20 Bedford way, London
WC1H 0AL, UK.
Disadvantaged children’s “low” educational expectations: Are the US and UK really so different to other industrialized nations?
John Jerrim∗†
Abstract. In most countries, children from disadvantaged backgrounds are
under-represented amongst the undergraduate population. One explanation
is that they do not see higher education as a realistic goal; that it is ‘not
for the likes of them’. In this paper, I use the Programme for International
Assessment data to investigate whether 15 year olds from disadvantaged
backgrounds are less likely to expect to complete university than their advantaged peers. I explore this issue across the OECD nations, though paying
particular attention to the US and UK. My results suggest that children from
less fortunate families are not as likely to make early plans for university as
their affluent peers. Yet the extent to which these findings differ across countries is rather modest, with little evidence to suggest that the UK stands out
from other members of the OECD. The US, on the other hand, appears to
be a nation where the relationship between socio-economic background and
the expectation of completing higher education is comparatively weak.
JEL classification: I21, I28, J62.
Keywords: Higher Education, University Access, Educational Expectations, PISA.
∗
Department of Quantitative Social Science, Institute of Education, University of London,
20 Bedford Way London, WC1H 0AL. E-mail: (J.Jerrim@ioe.ac.uk)
†
Thanks is given to the ESRC for providing funding for this project. I would particularly
like to thank John Micklewright, Patrick Sturgis, Sandra McNally and Sylke Schnepf for
their helpful comments, and the hospitality of the Institute for Research on Poverty at
the University of Wisconsin, Madison, on a visit while this paper was produced
5
“Children who grow up in inferior environments may expect less of themselves and may not
fully develop their academic potential because they see little hope for ever being able to
complete college or use their schooling in any effective way”
Cameron and Heckman (1999), Financing College Tuition: Government Policies &
Educational Priorities, page 76-124
Educational attainment has risen dramatically across the developed world over the past 15 to
20 years, with particularly strong growth in university participation. Yet despite this rising
trend, access to tertiary education remains unequal. Children with well-educated, affluent
parents are still over-represented in higher education, with relatively limited opportunities for
those from disadvantaged backgrounds. This issue has taken on particular prominence either
side of the Atlantic, where worries mount over equality of educational opportunity and social
mobility (see Blanden and Machin (2004), Machin and Vignoles (2004)). Consequently,
British and American governments have introduced policies to increase the number of
disadvantaged children entering higher education. Ensuring children hold high expectations
and aim for university from an early age is seen as a crucial step towards reaching this goal.
In other words, there is a belief that future educational plans made during adolescence have a
significant impact on later academic attainment, and a concern that poor children‟s low
expectations may be limiting their opportunities to succeed.
Initiatives to “raise expectations” have thus become a prominent feature of educational policy
in these countries, including the “Gear-up” and “I have a dream” programmes in the US and
the “Aim Higher” and “Gifted and Talented” schemes in the UK. It has also become a hot
topic of academic debate, with an increasing number of studies by economists (Chowdry et al
2009, Chevalier et al 2009 and Emmerson et al 2005), sociologists (Reynolds and Pemberton
2001, Morgan 2005) and social psychologists (Schoon 2010, Patton and Creed 2007,
Gottfredson 2002), all of which investigate adolescents‟ educational expectations within a
single national setting (typically either the US or UK). I contribute to this literature by
placing the link between family background and children‟s educational expectations (which is
often the focus of such studies) into a comparative context. Specifically, this paper shall
investigate whether the socio-economic gap in children‟s educational expectations is
particularly 'big' in the UK and US compared to other developed nations, and whether this is
greater than one would anticipate given these countries level of educational inequality.
6
I begin in section 2 by reviewing the relevant literature. This includes a discussion of how
children's expectations differ to their aspirations, and how this concept is linked to their
eventual educational attainment. Section 3 describes the Programme for International Student
Assessment (PISA) data on 15 year old children that I analyse and my empirical
methodology. During sections 4 and 5 I discuss the results. I conclude in section 6 with a
discussion of how my findings may inform educational policy in countries that encourage
disadvantaged children to 'aim higher', like the UK.
2. Existing literature and research questions
Friedman and Friedman (1980), amongst others, have argued that all young adults who can
benefit from university should have access to the returns it offers, regardless of their family
background. One reason is that this may lead to a more equitable society. Yet it is also
important for economic efficiency. Labour is a scarce resource that needs to be allocated
appropriately, but the brightest children may be excluded from the best jobs if they are unable
to 'fully develop their academic potential'. However, as noted in the introduction,
disadvantaged children tend to be under-represented amongst the undergraduate population
and in the most prestigious jobs (see Sutton Trust 2009). As suggested by Cameron and
Heckman (1999) at the start of this paper, some disadvantaged children may perceive there to
be a lack of opportunity to complete higher education which stops them from applying or, in
the words of Shields and Mohan (2008), believe that university is „not for the likes of them‟ 1.
Consequently, policymakers in the US and UK have introduced a series of programmes to
raise disadvantaged children‟s expectations of being able to obtain a tertiary level
qualification.
At this point, it is important to draw a distinction between children's “expectations” and their
“aspirations”. The former implies a realistic assessment of future outcomes, while the latter
reflects children‟s hopes and dreams (Gutman and Akerman 2008). So if a child expects to
obtain a university qualification, they truly believe that they will go on to complete this level
1
For instance a report by the Sutton Trust (2008), a UK based charity, states: 'exam grades on their own will not
necessarily lead to university if young people do not have a high level of expectation and make ill-informed
decisions'
7
of education. It is this concept that I attempt to explore in this paper. However, one must
consider whether 15 year old children (the age group that I study) are able to make such
realistic assessments of the future. Drawing from the developmental literature, Gottfredson
(2002) notes that, around age 14, children are beginning to recognize the need for
compromise in their educational and occupational goals. Likewise, Gutman and Akerman
(2008) suggest that at this age young people 'relinquish their most preferred choices and settle
for more acceptable, available choices', recognising the external constraints that they face.
From a different prospective, Morgan (1998) finds that adolescents' educational expectations
are not 'irrational fantasies'; rather, they are grounded in logical thinking, and vary with the
marginal costs and benefits associated with such continued schooling. Hence there is
evidence which suggests that young adults are able to distinguish between their aspirations
and expectations. Accurately capturing such details in a social survey is, however, another
matter. I shall further elaborate on this point when discussing the PISA data in the following
section.
It is also important to make clear that the value of any scheme that attempts to 'raise
disadvantaged children's expectations' is based on the assumption that this will have a causal
influence on their later behaviour and attainment. A conceptual model to illustrate this
relationship is set out in Figure 1, drawing upon the work of Chowdry et al (2009).
Figure 1 about here
This framework recognizes the multi-dimensional nature of family background, based around
measures of parental education, occupation (socio-economic status) and income. The authors
then specify four “transition mechanisms” (schools, neighbourhoods, parental attitudes,
material resources) by which family background influences children‟s attitudes, behaviour,
beliefs (including their aspirations for the future) and outcomes at age 14. The main focus of
this paper is, however, on the next stage of the model – the transition from adolescence to
young adulthood (i.e. from age 14 to 18). During this period, children may change their
attitudes, behaviours and beliefs about the future, which, in turn, alters their academic
trajectory2. Based on the work of Gottfredson (2002), I propose that one key development
between these ages is that children begin to recognize the external constraints that they face,
2
This framework also recognises that family circumstances and parental characteristics will continue to play a
role.
8
and thus start to develop expectations about their future (regarding, in particular, higher
education). These expectations then become the key behavioral “transmission mechanism”
that encourage greater effort and investment in school and less “risky” behaviour (drinking,
drug use and early sexual activity) between ages 14 and 18 which, in turn, leads to higher
educational attainment.
It is important to recognise, however, that this is not a static relationship; children will
continually revise these expectations, based on their on-going attainment. Indeed, it is likely
that higher expectations lead to higher attainment, which leads to continued high
expectations, and so forth. Yet, as one can not identify the exact age at which such feedback
begins, it has proven to be methodologically challenging to estimate the extent to which one
factor is driving the other. Nevertheless, several authors have explored the association
between these variables, with some attempt to address the direction of causality. For instance,
Khoo and Ainley (2005) investigate the educational plans and achievement of a sample of
Australian teenagers. Estimating a structural equation model, they show that children's
expectations are strongly associated with their later outcomes, even after controlling for a
host of potentially confounding factors. In a similar manner, Reynolds and Pemberton (2001)
find that expecting to go to university at age 15 is almost a prequisite for actual later
attendance in the US; they show that less than 3% of children who do not expect to go to
university actually obtain a degree by the time they turn 30. Likewise, Morgan (2004) uses a
regression based path analysis to investigate whether educational expectations held during the
mid-teens determines entry into post-compulsory schooling in the US. In turn, he finds
evidence of a strong and statistically significant association. Of course, economists may
express concerns about the potential endogeneity of expectations in any regression based setup, particularly due to omitted variable bias. Consequently, Morgan (2004) shows that
expectations remain a highly significant predictor of later outcomes using an instrumental
variable analysis. However he also recognizes the difficulties of identifying such models in
this set-up. Brown et al (2004) use similar methods to Morgan, and find a strong and highly
significant relationship between children's expectations and later attainment in the UK. Using
panel data, with measurement of young adults' educational expectations at several ages,
Morgan (2005) finds university plans are serially correlated across time. He suggests that this
is consistent with an underlying dynamic causal relationship between expectations and
attainment as described above. Similarly, Chowdry et al (2009) find that a number of
disadvantaged children in England stop believing that they will enter university between ages
9
14 and 16, and that these teenagers subsequently make less academic progress than their
peers who maintain high expectations. In a wider context, Cowan (2009) investigates the
relationship between American teenagers' educational expectations and their chances of
engaging in risky behaviour. Using an instrumental variable analysis, he finds that
'anticipated schooling has an effect on behaviour above and beyond the effect of realized
schooling' and thus that raising children's expectations of completing university may prove to
be an inexpensive way of reducing their tobacco, marijuana and alcohol consumption.
Finally, one may have concerns about the negative consequences for young people if their
expectations are not met. Reynolds and Baird (2010), however, find no long term emotional
costs of “shooting for the stars”.
Given the above, any difference between advantaged and disadvantaged children's
educational expectations will lead to a division in their behaviour, attainment and eventual
graduation rates. Indeed, the framework set out in Figure 1 suggests that such a divergence in
beliefs may well occur; expectations are assumed to have six primary determinants (schools,
neighbourhoods, parental attitudes, family resources, childhood attitudes and prior
attainment) all of which are associated with family background. For instance, advantaged
children will tend to go to better schools, where teachers may build their children‟s academic
confidence and emphasise their ability to complete this level of study. Similarly, it may be
that only well-educated parents stress the wider benefits of learning (meeting new people,
broadening horizons, growing up) to their offspring, who become driven towards higher
education as a result. Availability of resources will also determine children‟s expectations;
those from less fortunate households may believe they are credit constrained and thus do not
have the necessary finance to complete higher education. There may also be peer and role
model influences, both in school and the wider community, where disadvantaged children do
not see university as a realistic goal because they do not know any adult who has completed
higher education and have few friends who believe they can achieve the same. Attitudes may
also be transferred between generations, such as ambition and work ethic, which could
influence children‟s educational plans via the extent they are willing to stretch themselves in
the future. Finally, as expectations involve the recognition of external constraints, they will be
tempered by children‟s pre-existing skill, with large socio-economic differences in academic
achievement already evident at age 14 (see Hanushek and Woessmann (2010) for a survey of
the international evidence).
10
The analysis I undertake in this paper is motivated by the theoretical framework and
empirical analysis described above, which suggests that children‟s expectations have an
important influence on their later academic attainment, and that there are likely to be large
differences in these expectations between socio-economic groups. For instance, the work of
Chowdry et al (2010) suggests that expectations regarding higher education differ
substantially between advantaged and disadvantaged teenagers in the UK, and that this makes
the biggest contribution to the widening of the socio-economic attainment gap towards the
end of compulsory schooling. However, few have considered whether the difference in
educational expectations between rich and poor is bigger in some countries than in others. Yet
there are numerous reasons to suspect that this may be the case. For instance, the paragraph
above described how family background is linked to children‟s expectations (e.g. through
schools, resources, parental attitudes etc). Yet the strength of such associations are likely to
differ across nations. For instance, the US and UK are known for having high levels of
income inequality (see OECD 2008) and quite segregated schooling systems compared to
other nations (see Jenkins et al 2006). Similarly, both have comparatively high costs of
university tuition, which may mean disadvantaged British and American teenagers are more
likely to feel credit constrained than their peers in other countries. Such factors may
consequently lead to these countries experiencing particularly large gaps in educational
expectations between socio-economic groups.
In my first research question I consider this issue, exploring the size of the socio-economic
gap in children‟s educational expectations across the OECD, with a focus on results for the
US and UK. Although concern has been expressed about the difference between advantaged
and disadvantaged children's educational expectations in these countries, I have never seen it
put into a comparative perspective. It is therefore difficult to know if the socio-economic gap
in expectations is especially 'big' within these countries, and whether this is a bigger
“problem” here than other parts of the developed world. Hence I ask:
Research Question 1. What is the absolute size of the gap between advantaged and
disadvantaged children's expectations of completing university? Is this gap particularly
large in the US and UK compared to other me mbers of the OECD?
11
Of course, these countries also differ in terms of educational inequality; the gap between rich
and poor teenagers test scores is greater in some countries than in others. Schutz et al (2008),
for instance, compares the relationship between family background and children‟s test scores
across a range of developed countries. They show that the association is particularly strong in
England, Scotland and the US compared to elsewhere. In other words, these countries seem
to suffer high levels of educational inequality, and hence one might also expect there to be
especially large socio-economic differences in teenagers‟ educational plans. My second
research question considers this possibility and explores whether countries with a big richpoor gap in teenagers test scores are also the ones with a big rich-poor gap in children‟s
educational plans. One particular focus of this analysis will be whether the expectation gap in
the US and UK is bigger than one would predict given their level of educational inequality, or
if these countries manage to „buck the trend‟ (i.e achieve a smaller gap in young peoples‟
educational expectations than one would predict given their level of educational inequality).
In summary:
RQ 2. Is the socio-economic gap in children’s educational expectations particularly big
in the UK and US, given their relatively high levels of educational inequality?
Indeed, given the arguments made above, one might argue that differences in test scores at
age 15 are entirely responsible for the socio-economic gap in children‟s educational
expectations; the only reason why advantaged 15 year olds are more likely to expect entry
into university than their disadvantaged peers is that they have developed superior cognitive
skills (e.g. “outcomes at age 14” in Figure 1) by this point in time. On the other hand,
sizeable differences may remain even after controlling for academic skill measured at age 15.
That is to say that disadvantaged 15 year olds may be less likely to expect a university
education than their wealthy peers, even if they score equally well on assessments nearing the
end of compulsory schooling. I again consider whether this is a specific problem to the US
and UK, or if the situation here is comparable to other parts of the developed world. My final
research question is therefore:
Research Question 3. Do the higher educational expectations of advantaged children
only reflect their higher test scores at age 15? After controlling for this factor, is the
socio-economic gap in the UK and US particularly large in comparison to other
me mbers of the OECD?
12
In answering the question above, I am not able to make a causal statement about the
relationship between children's test scores, socio-economic status and their expectations. As
laid out in Figure 1, children are assumed to begin making firm educational plans between
ages 14 and 16, yet the exact point in time is almost impossible to identify. It could be that
children start thinking seriously about university from a younger age than I can measure (e.g.
14), which has already had an impact upon their motivation at school, and is thus reflected in
their scores on the PISA test (taken at age 15) 3. In other words, the process of educational
expectations influencing motivation and behaviour has already begun, causing age 15 test
scores to be endogenous in the models that I estimate. This may be a particularly big issue in
countries like England where children have to make educational decisions at a young age,
and who receive regular updates on their ability through test performance. Hence this set of
results needs to be treated with caution, and interpreted simply as the socio-economic gap in
plans to enter higher education amongst children who manage to score the same on the PISA
tests.
To summerise, this paper has one central aim – to place the socio-economic gap in children‟s
higher educational plans in the US and UK into a comparative context. In other words, is the
difference between “advantaged” and “disadvantaged” children‟s educational expectations
greater in these countries than elsewhere? Section 3 now turns to the PISA data that I use to
explore this topic, with the results from my analysis to follow in sections 4 and 5.
3. Data
The data I use are drawn from the 2003 round of the Programme for International Student
Assessment (PISA); a study of 15 year-olds‟ cognitive skills held every three years. Although
46 countries took part, I restrict my analysis to 33 industralised nations 4. In each country, a
minimum of 150 schools were included in the sample, selected with probability proportional
to size. Thirty students were then randomly selected from within. Average response rates of
both schools (90%) and pupils (90%) were high, though this varies moderately between
countries5 . Further details are available in the PISA 2003 technical report (OECD 2004b). A
3
In other words expectations at prior time points (that I am unable to control for) are confounding the
relationships that I estimate.
4
Here I treat the constituent parts of the United Kingdom (England, Scotland and Northern Ireland) as separate
countries. Likewise, I separate Flemish from French Belgium.
5
The lowest of which was England, at 64% for schools and 77% for pupils . Micklewright et al (2010)
13
set of sampling weights are also provided by the survey organisers tha t tries to correct for the
unit non-response. The achieved sample size, across the 33 countries I consider, is 224,094.
As part of the study, children were asked to complete a questionnaire. This included the
question:
“Which of the following do you expect to complete” [emphasis in original question]
Lower secondary education (Middle or junior high school)
Upper Secondary education (High school)
Post-secondary non-tertiary (Vocational/technical certificate after high school)
Tertiary “Type b” education (Associate‟s degree)
Tertiary “Type a” education of higher (Bachelors degree or higher)
Country specific options were provided in the questionnaire. The phrases in brackets
illustrate these for the US. The primary outcome I analyse in this paper is whether the child
ticked the top category (bachelors degree or higher). Response rates to this question were
very high. Table 1 shows that almost 99% of children responded, from a low of 93% in
France to a high of 100% in Poland. Consequently, I exclude the few (1%) observations
where educational expectations are missing 6.
Table 1 about here
Recall that my concern in this paper is children's expectations (realistic assessments of their
future), not their aspirations (idealistic goals). As noted in section 2, the developmental
literature suggests that by the time of the PISA study (approaching age 16), children typically
separate one concept from the other. Indeed, there has been work in the sociological literature
that compares children's expectations to their aspirations around this age (see Patton and
Creed (2007)). Such studies usually distinguish between the two concepts by altering and
emphasizing the operative word (e.g. asking children what they would “like” to do, and then
what they “expect”). Yet, to my knowledge, there has been little work on the validation of
investigate this non-response, and create an alternative set of responses weights (as opposed to those provided in
the dataset by the survey organisers) to try and correct for bias in the estimates. They show that the UK only
moves one place in the PISA ranking of children‟s test scores once these weights have been applied.
6
These observations are not a random selection from the population. Rather they tend to be children with lower
test scores, who also do not have complete information on family background.
14
such questions in quantitative surveys. In particular, there seems scant evidence of whether
such subtle phrases are able to elicit the appropriate information from respondents.
Unfortunately, the question asked in PISA shares much of the same criticism. It emphasizes
the word “expect” using bold, underlined letters, yet provides children with no further
instruction. Hence this is the only guide they have towards reporting their expectations rather
than their aspirations. Whether such subtle wording can be adequately translated into other
languages, as required in this cross-national analysis, is a further concern.
If this question is actually capturing children's aspirations, then the proportion reporting that
they “expect” to complete university will be significantly higher than current graduation
rates7. If this only occurs in certain countries, then these nations will out-lie from the rest.
Indeed, if it is a translational issue that is causing this problem, language will be a common
theme amongst these outlying nations. I search for such patterns in Figure 2. Specifically, in
each country I compare the proportion of children who expect to obtain a degree (that I have
calculated from the OECD PISA data) with actual graduation rates drawn from OECD
(2009). The 45 degree line is where the proportion of children expecting to complete
university equals actual graduation rates.
Generally, Figure 2 suggests that responses are not out of touch with reality; 45% of OECD
children expect to complete university against actual graduation rates of 40%. Indeed, several
countries, including England, sit below the 45 degree line; the proportion of children
expecting to enter university is below actual completion rates. I do note, however, that there
are some countries where one may have concerns. For instance the proportion of Canadian
and American children “expecting” to obtain a bachelor's degree is significantly higher than
actual graduation rates. However, Reynolds and Pemberton (2001) point out that there are
high drop out rates from university (at least in the US), and as such the proportion of US
children expecting to complete university are at least in-line with current entrance rates8. It is
also interesting to see that the proportion of children expecting to complete university varies
quite substantially across the English speaking countries, suggesting that this cross-national
7
Of course, such a finding may just reflect that children are not very good at predicting the future (the question
does capture children‟s expectations, it is just that these expectations are inaccurate). Nevertheless, if this pattern
occurs consistently across all nations then one may question whether this is actually capturing adolescents‟
expectations
8
Reynolds and Pemberton (2001) noted a similar finding when using the American NLSY 97 sample. OECD
(2009) suggests that 65% of US school leavers enter university, very similar to the number I find expecting to
obtain a degree. Unfortunately this information is not available for Canada.
15
variation is not simply due to a difference in language. Another feature of Figure 2 is the low
correlation between the proportion of young people expecting to obtain a degree and current
graduation rates (r = 0.02). Hence there is no suggestion that countries with a greater
proportion of the population completing university are also the ones where children are more
likely to expect to complete higher education.
Figure 2 about here
Nevertheless, despite concerns with some countries, the overall pattern of response is quite
encouraging, and generally seems to be consistent with a measure of children's expectations.
Hence these data do seem to be of value in answering the research questions I set out in
section 2. Yet I am unable to investigate (and thus rule out) other potential problems
regarding measurement error. For instance, it might be that advantaged children have a
tendency to report their expectations and disadvantaged children their aspirations, and that
this particular response pattern varies across OECD countries. Likewise, I advise caution in
interpreting results for less developed members of the OECD like Turkey, Greece and
Mexico, where “expectations” often seem to be out of touch with reality.
I now turn my attention to the variables that I use to distinguish between children from
“advantaged” and “disadvantaged” backgrounds. In this paper, I view such concepts as multidimensional, reflecting a whole host of factors that one can not capture in a single variable
(e.g. income). Rather, I define “advantaged” and “disadvantaged” on the basis of three factors
– parental education, parental occupation and the home learning environment –following the
theoretical framework of Chowdry et al (2009) set out in Figure 1.
I start by describing the importance (and measurement) of parental education. As noted in
section 2, parents with more schooling might place greater emphasis on their children's
education, or instill a taste for learning in their off-spring. Likewise, parents may be able to
provide more information and encouragement about going to university if they hold a tertiary
qualification themselves, and perhaps act as educational role models. Parental education will
also be a key factor driving household income and children‟s cognitive development. This is
therefore a key distinguishing feature between children from advantaged and disadvantaged
backgrounds.
16
Information on parental education was captured through the sampled children as part of the
PISA background questionnaire. Specifically, children were asked to report the level of
education their mother and father completed at school and what type of tertiary qualifications
they hold 9. Schlutz (2005) and Jerrim and Micklewright (forthcoming) investigate possible
measurement error in such reports using the PISA 2006 wave, where parents and children
were asked separately to report mother‟s and father‟s level of education 10. They gave the
same category in two thirds of cases, though this was notably higher (around 86% of
occasions) when the parents held a degree.
These responses were then recoded by the survey organisers into ISCED levels of education,
a measure designed by UNESCO to aid cross-national analysis (though some differences in
definitions across countries may remain - Steedman (2001)). The highest ISCED level
achieved by either parent is then used to create the “highest parental education” variable.
Appendix Table 1 shows how this is distributed across each of the OECD nations, including a
“missing” category where this information is unavailable (typically 5-10% of cases)11.
As sample sizes become small for certain groups, I recode this information into three broad
categories (low, medium and high) following a similar re-categorisation in the Luxemburg
Income Study (a well known dataset often used in cross-national research). I define high as
holding a tertiary qualification (ISCED 5B or 5A+), medium as post-secondary schooling but
no experience of higher education (ISCED 4) and low as completed secondary schooling or
less (ISCED 0-3). This broad categorisation also helps to ensure that I have a sufficient
number of observations within the “advantaged” and “disadvantaged” groups that I define
later in this section.
9
Note that children were instructed to report this information for their mother and father like figures.
Consequently, children living in a household with a complex family structure, for instance with a step-mother or
step-father, may not be reporting the education of their biological parents. I have experimented with including a
variable that captures this in my analysis. However, I have chosen to exclude it as the estimated effect was
usually small and statistically insignificant.
10
The parental questionnaire that contains this information was an “international option” in 2006. This
information is therefore only available for 11 countries, and has relatively high rates of non-response. I could
not use the 2006 data for this analysis as it did not contain a question on children‟s educational expectations for
the US or UK.
11
These children are not a random sample from the population. Rather, they disproportionately come from
children who performed poorly on the PISA test and come from less well-off families. One may argue that this
could be driving some of the cross-national differences observed. Given the relatively small non-response in the
majority of countries, I do not attempt any correction for this issue. However, I do include a “missing” dummy
variable in all subsequent regressions to ensure these children are not dropped from the analysis.
17
The second measure of family background that I use to distinguish between advantaged and
disadvantaged children is parental occupation. This variable is probably the best proxy
available for household income and financial resources (which are unfortunately not collected
as part of the PISA study) that play an important role in the development of children's
educational expectations as laid out in section 2. Parental occupation will also pick up
relevant aspects of social class, such as the societies, cultures and communities the child has
grown up in.
As with parental education, information on mother's and father's occupation was collected
directly from the sampled children. Specifically, they were asked the title of their mother's
and father's main job and a description of the type of work this involves. Responses were
coded by the survey organisers into four digit ISCO codes (the International Labour
Organisation's occupational classification), which assigns the reported occupation to one of
over 300 categories. Schulz (2005) investigates the potential measurement error in this data
using the 2006 PISA field trial. He found that parents and children reported the same
occupational group (defined in terms of the major ISCO groups) on roughly seven out of ten
occasions. My experimentations with the final 2006 PISA sample revealed similar results.
Children‟s responses were then coded by the PISA survey organisers into the quasicontinuous ISEI index of occupational status, designed by Ganzeboom et al (1992). This
index assigns each occupation a score between 16 and 90, depending upon the relevant
“inputs” (educational level required) and “outputs” (the salary commanded) from that
particular job12. Hence this is an objective occupational scale which is designed to be
correlated with income. Moreover, Ganzeboom et al (1992) specifically designed this scale to
aid the type of cross-national analysis I undertake in this paper, and have thus attempted to
validate it as a measure of socio-economic status across a large number of developed
countries (although some still question aspects of its validity – see Bukodi, Dex and
Goldthorpe (forthcoming)). Nevertheless, the ISEI index remains an attractive measure
against the possible alternatives (such as an aggregation of the 4 digit ISCO codes into the 9
major occupational groups). Summary statistics for the distribution of this variable across the
12
The OECD describes: “The index captures the attributes of occupations that convert parents‟ education into
income. The index was derived by the optimal scaling of occupation groups to maximise the indirect effect of
education on income through occupation and to minimise the direct effect of education on income, net of
occupation (both effects being net of age).”
18
OECD countries can be found in Appendix Table 2. It is interesting to note that the
distribution of the ISEI index generally seems to be quite similar across countries, with very
little cross-national variation in the 10th and 90th percentiles, and only slightly more at the 25 th
and 75th percentiles.
The third variable that I use to measure family background is children's reports of the number
of books at home. It has been argued that this is correlated with a number of aspects of family
background including parental education, household income and social origin
(Ammermueller and Pischke 2009, Schuetz et al 2008). Yet the same authors also suggest that
it picks up factors like the value parents place on their children‟s education and the
encouragement they provide with regards to schooling. Likewise, the PISA survey organisers
argue it is a measure of the 'home educational resources' available to the child. Hence this
picks up such residual aspects of family background that are not fully captured within my
measures of parental education and occupation, but are nevertheless likely to be important in
the development of children's expectations.
This information was also reported by the participating children in the background
questionnaire. Specifically, they were asked 'how many books are there in your home'
(excluding magazines, newspapers and textbooks) with six possible options. However, the
two bottom and two top categories contain a rather sparse number of observations. Thus I
combine the bottom (0-10, 11-25, 26-100), and top (101-200, 201-500, above 500) three
fields to form low and high groups, along with a 'missing' category, following a strategy
similar to that used by the survey organisers (see OECD 2004c page 283). As with my recategorisation of parental education, this also helps to ensure that I have sufficient
observations within the “advantaged” and “disadvantaged” groups that I shall define in the
following paragraph. The distribution of this variable across the OECD countries can be
found in Appendix Table 3.
In the next section, I use the aforementioned variables in a logistic regression model of
children's educational expectations. In all models, I also control for gender and whether the
child was a first or second generation immigrant (as this group may be under different
pressure from their family to complete higher education) 13. Likewise, in all estimations I
13
Children had to answer three questions regarding whether they, their mother or their father was born outside
the country that they are taking the test in. I define a child as an “immigrant” if they answer yes to any of these
19
include 'missing' categories (dummy variables) to ensure children are not dropped from the
analysis when pieces of information are unavailable. Thus the final form of this model is:
  ( E ij ) 
     1 .Sex i   2 .I i  β 3 .SES i  β 4 .I i * SES i
log
 1  ( E ij ) 


Where:
 ( Eij ) = Probability of the child expecting to graduate from university, where
E= 1 if the child expects to complete university, 0 otherwise
Sex = A binary indicator of the child‟s gender (0 = female, 1 =male).
I= Whether the child is a first or second generation immigrant (0 = Native , 1 = Immigrant)
SES = A vector of variables capturing the child‟s socio-economic background. This includes:

Highest parental education – A set of two dummy variables, one referring to “some
post-secondary education but no tertiary” (medium) the other “tertiary and above”
(high). (Ref: “compulsory schooling or less” – i.e. low)

Number of books in the home - A single dummy variable referring to whether there
are “more than 100 books” (high) in the family home (Ref: “Less than 100 books” low)

Highest parental occupation measured on the ISEI scale - Entered as a piecewise
linear term with knots at the 10 th, 25th, 50th, 75th and 90th percentile of the national
ISEI distribution
I then use this model to generate predictions of how likely a hypothetical child with given
characteristics is to expect to complete university. Specifically, I create these predictions for:
three questions.
20
1. An “advantaged” child
Defined as:

Either of their parents holds a tertiary qualification (“high” parental education)

There are over 100 books in the family home (“high” books)

The highest occupation of their parents sits at 75th percentile of the national ISEI
distribution

Country native

Female
2. A “disadvantaged” child
Who I define as:

Neither parent has completed any post-compulsory schooling (“low” parental
education)

There are less than 100 books in the family home (“low” books)

The highest occupation of their parents sits at 25th percentile of the national ISEI
distribution

Country native

Female
I then calculate the difference between these two predictions in order to compare the
expectations of “advantaged” and “disadvantaged” groups 14.
Note that the definition of “advantage” and “disadvantage” that I use in this paper is multidimensional. Specifically, the prediction for the “advantaged” group refers to those children
with “multiple advantages”. So not only is at least one of these children‟s parents working in
a high-level occupation, but also they have good access to educational resources and at least
one of their mother or father is university educated. The second group can be thought of as
“multiply disadvantaged” (i.e. low parental occupation, low parental education, poor access
to home educational resources) following a similar logic. In section 5 I shall use an
alternative definition of family background, and of “advantaged” and “disadvantaged”
groups, to test the robustness of my results
14
See Appendix Table 4 for the proportion of children who are defined as “advantaged” and “disadvantaged”
using this definition
21
The final variable that needs to be described, which forms an integral part of my second and
third research questions, is the PISA measure of children's academic skill. As part of the PISA
2003 study, children (aged 15) sat a two hour test. The PISA consortia claim that this
measures children's 'functional ability' (how well they can use the concepts examined in 'real
life' situations) in three domains (reading, maths and science). In 2003, maths was assigned as
the major domain, where the vast majority of questions children were asked were on this
topic. All test questions were explicitly designed with cross-national comparability in mind.
Answers were summarized by the survey organizers into a single score for each of the three
domains using an „item-response model‟; the intuition being that true skill in each subject is
unobserved, and must be estimated from the answers to the test. Consequently, five „plausible
values‟ are generated for each pupil, estimating their true proficiency in each subject. These
scores were scaled by the survey organizers to have a mean (across all OECD countries) of
500 points and standard deviation of 100. Throughout my analysis, I use the first of these
plausible values for the maths domain15. The correlation between test scales is high (r≈0.8
between maths and reading, and the same between maths and science), with little change in
my substantive results when I use the reading and science scales instead16. The distribution of
this variable across all the countries I consider can be found in OECD (2004a).
Throughout this paper, I shall present results mainly in terms of log-odds. This measure is
more attractive than alternatives like the odds ratio and marginal effect (predicted
probabilities) as they are not sensitive to the point on the logistic distribution on which they
are estimated, and are therefore not influenced by differences between countries in the
absolute proportion of children who expect to complete higher education. I illustrate this
point in Table 2. The second and third columns present the proportion of children expecting
to complete university depending on whether either of their parents holds a degree. Column 4
provides the marginal effect (the percentage point difference between columns 2 and 3) while
column 5 illustrates the difference in terms of the log odds.
Table 2 about here
15
I experimented using the other plausible values, and by running five separate models and averaging the
estimated coefficients and standard errors. Results are very similar to those presented.
16
Note that only around half the children within each country actually answer questions in each of “minor”
PISA domains (reading and science). Scores are estimated by the study organisers for the remaining children
using a Rasch modelling approach.
22
Comparisons between countries can look very different depending on which measure is used.
Take England, one of my countries of interest, and Korea. The difference between the second
and third column is quite similar in terms of the log-odds (1.69 in Korea to 1.73 in England),
but very different when considering the marginal effect (22 percentage points compared to
39)17. This is being driven by the fact that, across the population, Korean children are
generally more optimistic about their prospects of completing university than those in
England (77% expect to obtain a degree in the former, compared to 29% in the latter). As my
concern in this paper is the expectations of disadvantaged children relative to their
advantaged peers, I prefer the log-odds as it abstracts from the absolute proportion of the
population believing that they will complete higher education. However, appreciating that
this metric is rather cumbersome to interpret, I also occasionally present predicted
probabilities to assist the reader‟s understanding.
4. Results
I shall now present results for the research questions set out in section 2. Parameter estimates
and an illustration of my predictions for England and the US can be found in Appendix 1,
with those for other countries available upon request.
Figure 3 illustrates the gap between advantaged and disadvantaged children's expectations of
completing higher education18. In all countries, disadvantaged children are less likely to
expect entry into university than their more affluent peers. This gap is generally big (around
two and two and a half log odds in most countries) and is always significantly different from
zero at the one percent level. To put this into perspective, if a hypothetical disadvantaged
child had a 50% chance of expecting to complete university, the probability for their identical
(but advantaged) peer would be closer to 90%19.
Figure 3 about here
17
Across all countries, the estimated correlation between the marginal effect and log odds is 0.78.
A list of country abbreviations can be found in the left hand column of Table 1.
19
These probabilities were calculated using the formula: probability = exp(log[odds]) /(1+exp(log[odds])) . Log
odds of 0 correspond to a probability of 50%. Log odds of 2.5correspond to a probability of 92%. Hence, in this
hypothetical example, a difference of 2.5 log odds (i.e. the difference between advantaged and disadvantaged
groups) leads to a 42% difference in the probability.
18
23
Focusing firstly on the results for England, the difference in log odds between advantaged
and disadvantaged children's expectations is roughly 2.4; the 10th largest estimate in the
OECD. Larger gaps are found in Germany, Switzerland and Austria; countries where access
to university is restricted to children on the appropriate educational “track” 20. Dustman
(2004) shows that this “track” is strongly associated with family background, hence one
would anticipate there to be large differences between advantaged and disadvantaged
children's expectations in these countries. Likewise, in a number of Eastern European
countries (e.g. Hungary, Slovakia) the log-odds seems large when compared to other nations.
This should not been surprising, as several studies (see Shavit and Blossfeld 1993) have
shown that these countries have a large degree of educational polarisation. England is also
ranked higher than the other Anglo-Saxon countries. This includes Scotland and Northern
Ireland (the other constituent parts of the UK) where the estimated difference in log odds is
roughly equal to the OECD average (around 2.2). However the 95% confidence interval (the
thin black line running through the centre of each bar) suggests that caution is required when
interpreting this result. One can not reject the null hypothesis that England is significantly
different to Scotland or Northern Ireland at any of the conventional threshold. Indeed, Table 3
shows that the socio-economic gap in England is only significantly stronger than only two
countries at the 5% level (Finland and the US) and a further one at each of the 5% and 10%
levels (Turkey and Portugal). To summerise, there is little evidence that England stands out in
comparison to other parts of the UK (Scotland and Northern Ireland) or generally amongst
the OECD.
Indeed, the general suggestion of Figure 3 is that cross-national variation in the socioeconomic expectation gap is rather modest – although there are some interesting outliers.
The US, for instance, immediately stands out in Figure 3 - though perhaps not in the
direction one might initially expect. The socio-economic gap here on the log-odds scale is
around 1.6; the second smallest out of the 33 countries. Moreover, this result is not just a
matter of sampling variation; Table 3 shows that results for the US are statistically different to
11 other nations at the 1% level and a further 5 and 4 at the 5% and 10% levels. Hence it
seems that the absolute gap between advantaged and disadvantaged children's educational
expectations is actually quite “small” in the US – at least when compared to other members
of the OECD.
20
In these countries, children are sorted into different schools by their level of ability at a young age (known as
“tracking”).
24
Is this difference in educational expectations greater or smaller than we would expect given
each countries level of educational inequality? The answer to this question can be found in
Figure 4. Specifically, I re-estimate exactly the same model as set out in section 3 – but now
using Ordinary Least Squares (OLS) with children‟s age 15 test scores as the response. This
provides a measure of inequality in educational achievement that can be found on the x-axis.
The y-axis, on the other hand, provides the socio-economic gap in children‟s educational
expectations (as just presented in Figure 3). Running through the centre of the graph is a
regression line, which represents the difference in educational expectations one would predict
in a country given its level of educational inequality.
Figure 4
One of the most notable features of Figure 4 is the relatively strong correlation
(approximately 0.7) between educational inequality and the socio-economic gap in university
plans. This is in stark contrast to Figure 2 where I demonstrated that the correlation between
the proportion of children expecting to go to university and the proportion actually attended
was close to zero. In any case, Figure 4 indicates that England, Scotland and Northern Ireland
all sit very closely to the estimated regression line. Hence, despite policy concern in the UK
that disadvantaged children are a lot less likely to expect to complete higher education than
their advantaged peers, the international evidence suggests this is not particularly more than
one would anticipate given its level of educational inequality. The US, on the other hand,
again emerges as an intriguing outlier. It is the furthest below the regression line out of any of
the OECD countries. That is, the difference between advantaged and disadvantaged children‟s
educational expectations in the US is much lower than one would predict given its level of
educational inequality. One can, of course, come up with several possible explanations for
why this might be. For instance, the US is a country with a diverse ethnic composition and a
large number of young people of afro-Caribbean dissent. Such minority groups are known to
generally hold high educational expectations, but also suffer from a large “expectationattainment gap” (i.e. their expectations rarely match with reality - see Gutman and Akerman
for a discussion of this literature). Alternatively, one might argue that policies to raise
disadvantaged children‟s expectations of completing higher education are long established in
the US, and that the comparatively narrow gap between rich and poor is essentially
illustrating such initiatives success. It is beyond the scope of this paper, and indeed the data
available, to try and distinguish between these explanations. But attempting to explain why
25
such a pattern emerges in the US (and why it stands out from other countries) is an interesting
possibility for future research.
A final feature of Figure 4 that is worth highlighting is the positions taken up by countries
that “track” children into different schools from an early age. Note how these countries
(shown in circles) tend to sit above the line and mostly to the right (with the exception of
Switzerland). These are countries where educational inequalities tend to be high, but also
where the difference between advantaged and disadvantaged children‟s educational
expectations is greater than one might predict. One cautious interpretation of this result is
that, along with tracking being associated with educational ineq uality (as suggested in studies
by Hanushek and Woessmann 2006 and Ammermueller 2006), it may also be related to the
size of the socio-economic gap in adolescents‟ expectations for the future.
In Table 4, I investigate whether England and the US still stand out in the international
ranking of SES differences in educational expectations if I now control for differences in
children‟s age 15 test scores (i.e. when I add children‟s score on the PISA test as an additional
right hand side variable to the model that was presented in section 3). I of course recognise
the potential endogeniety of this variable, and that estimates from these models are not
causal; they should hence be treated simply as conditional associations found in the data. This
limitation does not, however, restrict my ability to address the simple question I am asking –
does there remain a socio-economic gap in educational expectations amongst children who
score the same on tests at age 15, and do the cross-national patterns remain similar to those I
presented before? Results can be found in Table 4.
Table 4 about here
The US is again placed towards the top of the table, while England is around the middle. The
socio-economic gap in the latter is now significantly stronger than in only one other OECD
countries at the five percent level (Finland), and a further two at the ten percent level (US and
Mexico). Once again, there is little difference compared to Scotland and Northern Ireland,
and all three countries sit broadly in line with the OECD average of around 1.7 log-odds.
Consequently, after controlling for test scores, there remains no evidence that the difference
between the educational expectations of advantaged and disadvantaged children is unusually
large in the UK. On the other hand, the association between family background and children's
26
expectations of completing university remains significantly weaker in the US than in other
OECD countries. Five countries are significantly different at the 1% level, with a further
seven at the 5% level and six more at the 10% level. This, once more, includes countries
within the United Kingdom, several from Scandinavia (e.g. Sweden, Norway) and those
which assign children to different educational tracks (e.g. Germany, Switzerland and Austria).
Hence, there remains a suggestion that the socio-economic gap in educational expectations is
comparatively small in the US.
5. Robustness of results
In this section, I use a different measure of family backgro und to test the robustness of the
aforementioned results. I do so for two reasons. Firstly, using my initial definition of
“advantage” and “disadvantage”, a different proportion of the population in each country is
being assigned to these groups. For example, Appendix Table 4 shows that 9% of English
children were defined as “disadvantaged” under the definition I described in section 3,
compared to 3% of those from Norway and roughly 20% in Portugal, Poland and Turkey.
Hence I have thus far presented results where a more extreme proportion of the population
has been defined as “advantaged” / “disadvantaged” in some countries than in others. I will
now re-perform my analysis, using a different measure of family background, which will
allow me to compare the most advantaged and the least advantaged quarter of the population
within each country (i.e. I will now investigate whether my results hold when defining the
same relative proportion of children in each country as “advantaged” and “disadvantaged”).
Secondly, the definitions and measures that I used in the previous section were just one (quite
specific) way to divide children into advantaged and disadvantaged groups. It is important to
test whether my results and substantive conclusions hold under possible alternatives.
To do so, I turn to the “Economic, Social and Cultural Status” (ESCS) index; a continuous
measure of family background (scaled to be mean 0 and standard deviation 1 across the
OECD countries) that is produced by the PISA survey organisers and is contained within the
provided dataset. Specifically, the survey organisers have produced a weighted average, via a
principal component analysis, of three variables (highest level of parental education, parental
occupation, and availability of items in the family home) to generate a measure of children‟s
socio-economic status. The first two of these variables (parental education and occupation)
are as described in the earlier data section (although parental education has now been
27
converted by the survey organiser into a linear term reflecting „years of schooling‟ – see
OECD 2004: 308 for details). The “availability of home possessions” is itself an index (from
another principal components analysis) based upon children's reports of whether they have
various items (e.g. computers, works of art, number of books) in their family home.
According to OECD (2004), this provides an approximate measure of household wealth.
Further details on this measure and its construction can be found in OECD (2004b), while
Schulz (2005) investigates its properties (reporting reasonable levels of internal consistency
and stability).
This index has several attractions as an alternative measure of family background. Firstly, it
continues to capture the multi-dimensional nature (education, occupation, income/wealth) of
“advantage”. Secondly, as this variable is continuous, I can easily widen the proportion of
children contained within my definition of advantaged and disadvantaged groups (to, for
instance, the top and bottom quartile). Also note that, by using this measure, I can ensure that
the same relative proportion of the population is defined as advantaged and disadvantaged in
each of the OECD nations. Yet this variable also has a number of limitations. As it is created
via a principal components analysis, it is somewhat difficult to interpret. There is also likely
to be some information loss from suppressing various measures into one, all-encompassing,
continuous index. One may also have some doubts over the validity of using household items
as a measure of family wealth. Whether a child grows up in a home with a dishwasher or
works of art will to some extent reflect parental preferences, and thus may provide little
insight into whether they truly come from an advantaged or disadvantaged background.
Similarly, one may question the cross-national comparability of such measures 21. Yet, despite
these limitations, this remains an attractive alternative measure of socio-economic status due
to its flexible nature. The distribution of this index across OECD countries can be found in
OECD (2004a).
21
Indeed, it is for these reasons that I did not use this variable or include such information in my initial definition
of advantaged and disadvantaged groups (described in section 3).
28
I proceed by dividing this variable into four equal groups (separately for each country) and
defining:
„Advantaged‟ = top quartile group of the national ESCS distribution
„Disadvantaged‟ = bottom quartile group of the national ESCS distribution
I then use this information in a regression model as a set of dummy variables, with the bottom
quartile („disadvantaged‟) as the reference group. Formally, this model is specified:
  ( E ij ) 
   1   1 .Sex i   2 .I i  β 3 .SESi  β 4 .I i *SES i
log
 1   ( E ij ) 


where:
SES = A vector of three dummy variables reflecting advantage, based upon quartiles the
ESCS measure of family background described above (reference = bottom quartile)
All other variables are as described in section 3.
The estimated coefficient for the “top quartile” dummy variable captures the difference
between “advantaged” and “disadvantaged” children‟s educational expectations, and can be
found on the y-axis of Figure 5. I then re-run the model above, but using OLS regression with
children‟s scores on the PISA maths test as the response. This then provides a measure of
educational inequality within each country, which can be found on the x-axis. One may think
of these results as analogous to those presented previously in Figure 4, but now using this
alternative definition of “advantaged” and “disadvantaged” groups.
Figure 5
Generally, these results are largely consistent with those that I presented before, although
there is slightly more variation around the estimated regression line (the correlation is now
roughly 0.55 down from 0.70 for the estimates presented in section 4). Nevertheless, England
and Scotland still sit very closely to this predicted line – again suggesting that the difference
in advantaged and disadvantaged children‟s educational expectations is not greater than one
29
would anticipate given their level of educational inequality. Likewise, it still seems to be the
case that countries which “track” children into different types of school from an early age
tend to sit above the 45 degree (although not by particularly large magnitudes in most cases).
The US, on the other hand, is slightly less of an outlier than before; it does not stick out quite
so much from all other developed countries. Yet, it is still quite some distance from the fitted
line, and there remains a strong suggestion that the socio-economic gap in children‟s
educational expectations is smaller than one would predict given the level of educational
inequality in the US. Hence the overall message is one that tally‟s with the preceding section,
with my substantive conclusions largely unchanged.
6. Summary and conclusions
There is a concern in many countries that children from disadvantaged backgrounds are
under-represented amongst the undergraduate population. In particular, policy-makers are
worried that some young adults who could benefit from higher education decide not to seek
out the returns that this investment can offer. One explanation for this is that many
disadvantaged young people in these countries see university as “not for the likes of them”.
As such, there has been increasing concern regarding the socio-economic gap in children‟s
expectations of completing higher education. “Widening access” schemes that try to address
this issue have thus become common across the developed world. A particular feature of such
programmes in the US and UK is that they explicitly aim to raise disadvantaged children's
expectations of completing university. It is claimed that, in raising the proportion of
disadvantaged students expecting to complete higher education, such policies will reduce the
socio-economic divide in tertiary graduation rates.
By considering the size of this expectation gap in the US and UK, and how it compares to
other developed countries, this paper makes an important contribution to the existing
literature. I have shown that, although there are large differences between advantaged and
disadvantaged children's educational plans, this holds true across all countries in the
developed world. There is little evidence that the UK stands out when compared to other
members of the OECD, and that the socio-economic gap between “advantaged” and
“disadvantaged” children‟s educational plans is roughly what one would predict given this
country‟s level of educational inequality. There is, on the other hand, quite a strong
suggestion that the socio-economic gap in educational expectations is atypically small in the
30
US.
It is of course important to note the limitations of this study and, in doing so, aspects of the
wider literature. To begin, I remind the reader that none of the estimates I have presented
should not be treated as causal. Rather this paper has simply attempted to give the socioeconomic gap in children‟s educational expectations, which is of great academic and political
interest in many countries, a comparative context. On a related issue, I have undertaken this
research based on the assumption that adolescents' educational expectations have an
important influence on their later behaviour and schooling attainment. Although there is
evidence supporting this from a broad range of disciplines, including sociology, social
psychology and economics, further work in this area still needs to be done. In particular,
future research should focus on untangling the relationship between these variables and
whether such associations vary across different national settings. Finally, all my analyses and
subsequent inferences are based on the assumption that children are reporting their
educational expectations, rather than their aspirations, in the PISA survey. Although general
patterns within the data are consistent with this view, formal validation of this type of
question would represent a significant forward step for the wider literature.
One should, nevertheless, not lose sight of the contribution this paper makes to the wider
literature. Single national studies are always likely to find a socio-economic gap in attitudes
and expectations towards higher education, producing differences that (at first sight) tend to
be rather striking. But before deciding whether this expectation gap is “big” or not, and if a
country has a particular “problem” in this area, it needs to be put into context. With reference
to the US and UK, although there is a significant socio-economic gap in children's
educational expectations, it is not atypically big compared to other developed nations. In fact,
given its level of educational inequality, the socio-economic gap in university expectations in
the US is rather small. This does not mean that policymakers should stop their investment in
this area. However, it is important that they understand that socio-economic differences in
attitudes and expectations regarding higher education is not an area where the UK is doing
particularly badly (at least when this data was collected in 2003)22. Similarly, for
22
It is worth bearing in mind that since this data was collected in 2003, higher education finance in the UK has
undergone a substantial change. It is difficult to say whether higher tuition fees (introduced in 2005 and to be
extended in 2012) may alter this result. For example, one could argue that higher tuition fees may make
disadvantaged children feel more credit constrained, hence lowering their expectations towards HE compared to
their advantaged peers. The UK has unfortunately not collected data on educational expectations in subsequent
31
policymakers and academics in the US, it is worth bearing in mind that getting disadvantaged
young people to aim high is something which they already do comparatively well.
PISA survey waves (i.e. past 2003), meaning I am unable to investigate this interesting possibility.
32
References:
Ammermueller, A. (2006) “Educational Opportunities and the Role of Institutions”, ZEW Discussion Paper No.
05-44, Mannheim.
Ammermueller A & Pischke J-S (2009) “Peer Effects in European Primary Schools: Evidence from the Progress
in International Reading Literacy Study”, Journal of Labor Economics , volume 27, p 315-48
Blanden, J. & Machin, S. (2004) “Educational Inequality and the Expansion of UK Higher Education”, Scottish
Journal of Political Economy, Special Issue on the Economics of Education, volume 51, p 230–49
Brown, S. Sessions, J. & Taylor, K. (2004) “What Will I Be When I Grow Pp? An Analysis of Childhood
Expectations and Career Outcomes”, Working Paper No. 05/2, Department of Economics, University of
Leicester
Bukodi, E. Dex, S. & Goldthorpe, J. (forthcoming) “The Conceptualisation and Measurement of Occupational
Hierarchies: A Review, a Proposal and Some Illustrative Analysis”
Cameron, S. & Heckman. J. (1999) "Can Tuition Combat Rising Wage Inequality?", In Kosters, M. (ed),
Financing College Tuition: Government Policies & Educational Priorities
Chevalier, A. Gibbons, S. Thorpe, A. Snell, M. & Hoskins, S. (2009) “Students' Academic Self-Perception”,
Economics of Education Review, volume 28, p 716-727
Chowdry, H. Crawford, C. Dearden, L. Goodman, A. & Vignoles, A. (2008) “Widening Participation in Higher
Education: Analysis using Linked Administrative Data”, Institute for Fiscal Studies report
Chowdry, H. Crawford, C. & Goodman, A. (2009) “Drivers and Barriers to Educational Success Evidence from
the Longitudinal Study of Young People in England”, DCSF research Report RR102
Cowan, B. (2009) “Forward-Thinking Teens: The Effects of College Costs on Adolescent Risky
Behavior”, Universiy of Madison, Wisconsin job market paper
Chowdry, H. Crawford, C. , Dearden L., Joyce, R., Sibieta, L., Sylva, K. Washbrook, E. (2010) “Poorer
children‟s education attainment: how important are attitudes and behaviour?”
Dustmann, C (2004) "Parental Background, Secondary School Track Choice, and Wages," Oxford Economic
Papers, Oxford University Press, volume 56, p 209-230
Emmerson, C. Frayne, C. McNally, S. & Silva, O. (2005) “Evaluation of Aimhigher: Excellence Challenge: The
Early Impact of Aimhigher: Excellence Challenge on Pre-16 Outcomes: An Economic Evaluation”, Department
for Education and Skills Research Report 652
Friedman, M. & Friedman, R. (1980) “Free to Choose”
Ganzeboom, H. De Graaf, P. & Treiman, D. (1992) "A Standard International Socio-Economic Index of
Occupational Status", Social Science Research , volume 21, p 1-56
Gottfredson, S. (2002) “Gottfredson‟s Theory of Circumscription, Compromise, and Self Creation”, In D.
Brown (Ed.) Career Choice and Development , 4th edition, p 85-148
Gutman, L. & Akerman, R. (2008) “Determinants of Aspirations”, Centre for Research on the Wider Benefits of
Learning Research Report 27
Hanushek, E. & Woessmann, L. (2006) “Does Educational Tracking Affect Performance and Inequality?
Differences in Differences Evidence across Countries”, Economic Journal, volume 116, pp C63-C76.
Hanushek, E. & Woessmann, L. (2010) “The Economics of International Differences in Educational
Achievement”, IZA Discussion Paper 4925
Jenkins, S. Micklewright, J. & Schnepf, S. (2006) “Social Segregation in Secondary Schools: How Does
England Compare with Other Countries?”, IZA Discussion Papers 1959, Institute for the Study of Labor (IZA).
33
Jerrim, J & Micklewright, J (forthcoming) “Children‟s Education and Parents‟ Socio-economic Status:
Distinguishing the Impact of Mothers and Fathers”
Khoo, S. & Ainley, J. (2005) “Attitudes, Intentions and Participation”. LSAY Research Report 41
Machin, S. & Vignoles, A. (2004) “Educational Inequality: The Widening Socioeconomic
Gap”, Fiscal Studies, volume 25, p 107–28
Micklewright, J. Schnepf, S. & Skinner, C. (2010) “Non-response Biases in Surveys of School Children: The
Case of the English PISA samples”, S3RI Applications & Policy Series working paper A10/01
Morgan, S. (1998) “Adolescent Educational Expectations. Rationalized, Fantasized, or Both?” Rationality and
Society, volume 10, p 131–162
Morgan, S. (2004) “Methodologist as Arbitrator: Five Methods for Black-White Differences in the Causal Effect
of Expectations on Attainment”, Sociological Methods and Research, volume 33, p 3–53
Morgan, S. (2005) “On the Edge of Commitment: Educational Attainment and Race in the United States”,
Stanford University Press
Organisation for Economic Co-operation and Development (2004a) “Learning for Tomorrow‟s World – First
Results from PISA 2003”, Paris: Organisation for Economic Co-operation and Development
Organisation for Economic Co-operation and Development (2004b) “PISA 2003 Technical Report”, Paris:
Organisation for Economic Co-operation and Development
Organisation for Economic Co-operation and Development (2004c) “PISA 2003 data analysis manuals for SAS
and SPSS users”, Paris: Organisation for Economic Co-operation and Development
Organisation for Economic Co-operation and Development (2008) “Growing unequal? : Income distribution and
poverty in OECD countries” , Paris: Organisation for Economic Co-operation and Development
Organisation for Economic Co-operation and Development (2009) “Education At a Glance 2009: OECD
Indicators”, Paris: Organisation for Economic Co-operation and Development
Patton, W. & Creed, P. (2007) “The Relationship between Career Variables and Occupational Aspirations and
Expectations for Australian High School Adolescents”, Journal of Career Development, volume 34, p 127 -148
Reynolds, J. & Pemberton, J. (2001) "Rising College Expectations among Youth in the United States: A
Comparison of 15 and 16 Year Olds in the 1979 and 1997 NLSY", Journal of Human Resources, volume 36, p
703-726
Reynolds, J. & Baird, C. (2010) “Is There a Downside to Shooting for the Stars? Unrealized Educational
Expectations and Symptoms of Depression”, American Sociological Review, volume 75, issue 1, pp 151–172
Schoon, I. (2010) “Planning for the future. Changing education expectations in three British cohorts”, Historical
Social Research, volume 35, issue 2, p 99-119
Shavit, Y. & Blossfeld, H. (1993) “Persistent Inequality: Changing Educational Attainment in Thirteen
Countries”, Boulder, Colorado: Westview Press
Schütz G., Ursprung, H. & Wössmann, L. (2008) “Education policy and equality of opportunity”, Kyklos,
volume 61, p 279–308
Schulz, W. (2005) “Measuring the Socio-economic Background of Students‟ Achievement in PISA 2000 and
PISA 2003”, Paper presented at the Annual American Educational Research Association (AERA), San Francisco
Shields, C. & Mohan, J. (2008) “High-Quality Education for All Students: Putting Social Justice at its Heart”,
Teacher Development, volume 12, p 289-300.
34
Steedman, H. & McIntosh, S. (2001) “Measuring Low Skills in Europe: How Useful is the ISCED
Framework?”, Oxford Economic Papers, volume 53, p 564-581.
Sutton Trust (2008), “Increasing Higher Education Participation Amongst Disadvantaged Young People and
Schools in Poor Communities”, Sutton Trust report to the National Council for Educational Excellence
Sutton Trust (2009) “Sutton Trust submission to the House of Commons Children, Schools and Families
Committee on Social Mobility and Education and Access to the Professions”
35
Table 1. Sample sizes and missing expectations data across the OECD countries
Country
Poland (POL)
Finland (FIN)
Italy (ITA)
Japan (JAP)
Korea (KOR)
Spain (ESP)
Turkey (TURK)
Hungary (HUN)
Slovakia (SLOV)
Greece (GRE)
Portugal (PORT)
Switzerland (SWZ)
Sweden (SWE)
Australia (AUS)
Mexico (MEX)
Denmark (DEN)
Scotland (SCO)
USA (USA)
Austria (AUT)
Iceland (ICE)
Ireland (IRE)
Luxembourg (LUX)
Northern Ireland (NI)
Belgium(French) (BELFREN)
Norway (NOR)
New Zealand (NZ)
Netherlands (NLD)
Belgium(Flemish) (BELFLEM)
England (ENG)
Czech Republic (CZE)
Germany (GER)
Canada (CAN)
France (FRA)
Starting
sample
size
Complete
educational
expectation data
% Missing
expectations
data
4,383
5,796
11,639
4,707
5,444
10,791
4,855
4,765
7,346
4,627
4,608
8,420
4,624
12,551
29,983
4,218
2,723
5,456
4,597
3,350
3,880
3,923
2,853
2,958
4,064
4,511
3,992
5,838
3,959
6,320
4,660
27,953
4,381
5,793
11,631
4,700
5,440
10,776
4,852
4,756
7,328
4,613
4,594
8,393
4,605
12,492
29,845
4,191
2,707
5,419
4,558
3,324
3,848
3,892
2,829
2,931
4,023
4,447
3,902
5,696
3,817
6,076
4,457
26,707
0.0
0.1
0.1
0.1
0.1
0.1
0.1
0.2
0.2
0.3
0.3
0.3
0.4
0.5
0.5
0.6
0.6
0.7
0.8
0.8
0.8
0.8
0.8
0.9
1.0
1.4
2.3
2.4
3.6
3.9
4.4
4.5
4,300
3,997
7.0
TOTAL
224,094
221,020
1.4
Notes:
1 Missing data refers to item non-response only. Details on unit non-response can be found in the OECD
(2004b) Technical Report.
2 Data sorted by the percentage of missing observations
36
Table 2. Children’s expectations of completing university, depending on whether either
of their parents’ holds a bachelor’s degree
% Expecting to
complete university if
either their mother or
father holds a degree
% Expecting to
complete university
if neither their
mother or father
holds a degree
Marginal
Effect
Difference in
log-odds
Mexico
Netherlands
Finland
Belgium(French)
France
Portugal
Sweden
Canada
Ireland
Greece
Norway
Italy
USA
Denmark
Scotland
Japan
Spain
Australia
Turkey
Belgium(Flemish)
Switzerland
New Zealand
Luxembourg
Slovakia
Northern Ireland
Iceland
Korea
England
Austria
Germany
Czech Republic
Poland
69
56
67
48
56
73
50
80
76
81
45
72
82
47
73
69
74
80
93
59
39
70
71
73
64
64
93
60
59
44
73
67
53
31
41
23
28
45
23
54
48
55
19
41
55
19
41
36
41
49
75
24
12
32
33
35
26
26
71
21
20
12
31
23
16
25
25
25
28
28
27
26
29
25
26
31
28
28
33
34
33
30
18
35
27
38
39
38
37
38
22
39
40
32
42
44
0.68
1.04
1.07
1.13
1.19
1.20
1.21
1.23
1.23
1.25
1.25
1.31
1.32
1.33
1.36
1.38
1.41
1.43
1.49
1.52
1.55
1.60
1.60
1.61
1.62
1.62
1.69
1.73
1.75
1.75
1.79
1.92
Hungary
87
41
45
2.26
OECD average
70
40
30
1.25
Notes:
1 The column labeled „marginal effect‟ illustrates the percentage point difference between children‟s
expectations. This is the difference between the first two columns. Conversely, the final column illustrates the
difference between the same figures, but in terms of the log-odds.
2 The final row, labeled OECD average, refers to when one combines data across all 33 OECD countries
considered.
3 Countries sorted by the difference in terms of log-odds
37
Table 3. Difference between the expectations of advantaged and disadvantaged children
Country
Log odds
Fin
USA
Turk
Port
Fra
Ita
Ice
Can
Gre
Lux
NZ
Nld
NI
Ire
Pol
Den
Swe
Sco
Nor
BelFren
Esp
Cze
Eng
Swz
Aus
Aut
BelFlem
Slov
Hun
Germany
Hungary
Japan
1.41
1.59
1.70
1.93
1.94
1.96
1.97
1.99
1.99
2.01
2.05
2.05
2.11
2.11
2.14
2.15
2.20
2.22
2.23
2.24
2.36
2.39
2.43
2.43
2.47
2.56
2.62
2.81
3.09
3.20
3.21
3.31
Sig diff from
Eng?
SE
0.15
0.18
0.25
0.14
0.23
0.16
0.18
0.13
0.23
0.31
0.25
0.25
0.26
0.17
0.17
0.26
0.20
0.30
0.24
0.32
0.14
0.18
0.25
0.36
0.22
0.28
0.23
0.19
0.23
0.39
0.37
0.00
***
***
**
*
*
*
*
***
Sig diff from USA?
*
*
**
**
*
**
*
**
*
***
***
***
**
***
***
***
***
***
***
***
***
38
Table 4. Difference between the expectations of advantaged and disadvantaged children,
after controlling for age 15 test scores
Country
Log odds
Fin
USA
Mex
Nld
Turk
Port
Fra
NI
NZ
Ice
Lux
Ire
Can
Den
Cze
Pol
Sco
Ita
Esp
BelFren
Nor
Gre
Eng
Aus
Swe
Aut
Slov
BelFlem
Ger
Swz
Hun
1.095
1.190
1.202
1.232
1.275
1.416
1.437
1.442
1.524
1.557
1.567
1.600
1.636
1.641
1.654
1.689
1.692
1.695
1.704
1.738
1.740
1.743
1.752
1.760
1.835
1.871
1.968
1.979
1.984
1.992
2.116
SE
Sig diff from Eng?
0.151
0.182
0.153
0.256
0.250
0.147
0.245
0.294
0.247
0.192
0.344
0.186
0.139
0.267
0.189
0.180
0.309
0.154
0.149
0.331
0.246
0.253
0.264
0.241
0.196
0.264
0.184
0.242
0.300
0.359
0.237
**
*
*
-
Sig diff from USA?
*
*
**
**
**
*
*
*
*
**
**
***
***
**
**
***
Notes:
1 The final two columns illustrate whether the estimated difference in log odds are significantly different to
those in England and the US. *, ** and *** indicate a statistically significant difference at the 10%, 5% and 1%
level.
2 Countries sorted by the difference in expectations of advantaged and disadvantaged groups
3 Statistical significance calculated using a two sample t-test assuming independent samples are drawn between
countries
39
Figure 1. A model linking family background to children’s aspirations, expectations and outcomes, based upon Chowdry et al (2009)
FAMILY BACKGROUND
Parental socioeconomic status
“TRANSMISSION MECHANISMS”
Other family
background and
demographics
Unobservable
characteristics
CHANGES IN
“TRANSMISSION
MECHANISMS”
Outcomes at age 14
Schools
Neighbourhoods
Parental
education
EARLY OUTCOMES
Parental attitudes,
behaviours and
beliefs
Material resources
diverted to education
Key Stage 3 results
Young people‟s
attitudes,
behaviours and
beliefs
(e.g. aspirations)
Early smoking,
drinking and
cannabis use
Changes in
parents‟
characteristics
(Income,
Attitudes and
Resources)
Truancy
Anti-social
behaviour
OUTCOMES
Outcomes at ages
16-18
Key Stage 4 results
Smoking, drinking
and cannabis use
Truancy
Development of
young people‟s
attitudes and
beliefs
(e.g. the
development of
expectations)
Anti-social
behaviour
40
60
80
Figure 2. Proportion of children expecting to obtain a degree versus actual graduation
rates
USA
Can
Gre
Aus
40
Hun
Ire
EspScot Jap Port
Ita
Cze Slov
Nld
Fin
NZ
Ice
Swe
Eng
Pol
Nor Den
Aut
20
NI
Ger
Swz
20
40
60
% of school leavers who obtain a degree (OECD)
80
Notes:
1 Data on the % of school leavers who obtain a degree (x-axis) has been drawn from OECD Education At A
Glance Report (2009), Table A3.2, page 74. This refers to net graduation rates (i.e. as the sum of age-specific
graduation rates). See Annex 1 of OECD (2009) for further details. Information on Mexico, Luxemburg, Korea,
France and Belgium not available in this data. Data on the proportion of children who expect to obtain a degree
(y-axis) is based on my calculations from the PISA 2003 data.
2 Data is only available for the UK as a whole (not separately for England, Scotland and Northern Ireland) in the
OECD (2009) report. Hence I use data on higher education participation for these countries Data taken from:
http://www.dcsf.gov.uk/rsgateway/DB/SFR/s000716/SFR10_2007v1.pdf for England
http://www.scotland.gov.uk/Publications/2009/11/20112425/4 for Scotland
http://www.delni.gov.uk/he-api0607.pdf for Northern Ireland
3 To calculate statistical significance I have compared the proportion of children expecting to complete
university (drawn from the PISA data) to the OECD „Education At A Glance‟ figures of actual graduation rates. I
assume the latter refer to the population, hence conduct a one sample test of the PISA figures against these
values.
4 Distance from the 45 degree line (where average expectations = actual graduation rates) are statistically
significant in all countries at the 1% level except Germany, Austria, Netherlands, Slovakia, Czech Republic and
Finland
41
Figure 3. Estimated difference between advantaged and disadvantaged children's plans
to complete higher education (based on model 1)
Japan
Hungary
Germany
Hun
Slov
Ger
BelFlem
Aut
Aus
Swz
Eng
Cze
Esp
BelFren
Nor
Sco
Swe
Den
Pol
Ire
NI
Nld
NZ
Lux
Gre
Can
Ice
Ita
Fra
Port
Turk
USA
Fin
0
1
2
Estimated difference in log odds
3
4
Notes:
1 Results are based upon predictions from the regression model that I describe in section 3. These predictions
are based upon measures of highest parental education, highest parental occupation and number of books in the
home. Other controls include gender, immigrant status and an interaction between immigrant status and the
three measures of advantage listed above (note that children‟s PISA test scores are NOT included as a
covariate).
2 The thick, solid bars represent the difference between advantaged and disadvantaged children‟s expectations
of completing university, as measured in log-odds. The thin black line at the ends of these bars illustrates the
95% confidence interval of this estimate.
3 Country names corresponding to abbreviations can be found in the first column of Table 1
42
Figure 4. Estimated difference between advantaged and disadvantaged children's plans
to complete higher education versus difference in their scores in the PISA test
3
Hun
Slov
Ger
2.5
Aut
Swz
Aus
Eng
Cze
Esp
Sco
Lux
Gre
Can
Ice
Ita Fra
Port
BelFren Nor
Swe
Pol
2
BelFlem
Ire
Den
NI
Nld
NZ
Turk
1.5
USA
Fin
60
70
80
90
Difference in test scores
100
110
Notes:
1 Results on the y-axis are based upon predictions from the logistic regression model of children‟s educational
expectations that I describe in section 3. Specifically, this is the difference in educational expectations between
advantaged and disadvantaged groups in terms of log-odds. These predictions are based upon measures of
highest parental education, highest parental occupation and number of books in the home. Appendix 1 provides
further details. Other controls include gender, immigrant status and an interaction between immigrant status and
the three measures of advantage listed above. Note that these results are drawn from the first model specification
and so do not include information on the children‟s PISA test scores as a right hand side variable.
2 Results on the x-axis are based upon predictions from an OLS regression model of children‟s age 15 PISA
maths test score. The covariates in this model are exactly the same as those that enter the logistic regression of
educational expectations described in note 1 above. Figures refer to the number of points difference between
advantaged and disadvantaged groups on the PISA test. For example, advantaged children score roughly 85
points more on average than their disadvantaged peers in the US (note: 100 PISA points =1standard deviation).
3 Country names corresponding to abbreviations can be found in the first column of Table 1
4 Squares refer to the countries of particular interest in this paper. Circles indicate tracking countries that sit
above the line of best fit
43
Figure 5. Robustness test using national ESCS quartiles
3.5
Hun
3
Swz
Gre
BelFlem
Slov
Ger
Pol
Lux
Cze
Aut
2.5
Esp
Sco
2
Can
Ita
Nor
Swe
Ire
Ice
BelFren
Eng
Port
Fra
NldAus
Den
NI
NZ
USA
1.5
Fin
60
80
100
Difference in test scores
120
140
Notes:
1 Results on the y-axis are based upon predictions from the logistic regression model of children‟s educational
expectations that I describe in section 5. Specifically, this is the difference in educational expectations between
advantaged and disadvantaged groups in terms of log-odds. These predictions are based upon the difference
between the top and bottom national quartile of the ESCS measure of family background. Other controls include
gender, immigrant status and an interaction between immigrant status and the three measures of advantage listed
above. Note that these results are drawn from the first model specification and so do not include information on
the children‟s PISA test scores as a right hand side variable.
2 Results on the x-axis are based upon predictions from an OLS regression model of children‟s age 15 PISA
maths test score. The covariates in this model are exactly the same as those that enter the logistic regression of
educational expectations described in note 1 above. Figures refer to the number of points difference between
advantaged and disadvantaged groups on the PISA test (e.g. advantaged children score roughly 85 points more
on average than their disadvantaged peers in the US). A 100 PISA points difference refers to a 1 international
standard deviation change.
3 Country names corresponding to abbreviations can be found in the first column of Table 1
4 Squares refer to the countries of particular interest in this paper. Circles indicate tracking countries that sit
above the line of best fit
44
Appendix 1. Calculation of socio-economic expectation gap in section 4
In section 4 I present the difference between the educational expectations of a hypothetical
“advantaged” and “disadvantaged” child (in terms of log-odds). Recall that these hypothetical
children differ in terms of highest parental education, highest parental occupation and number
of books in the home as described in section 3.
Parental education and books in the home are dummy variables, where the reference group
refers to characteristics of the disadvantaged child (less than 100 books, neither parent
completed any more then compulsory schooling). One must sum the relevant coefficients
(those on the “high” books and “high” education dummies) to form this part of the prediction
of the difference between the hypothetical advantaged and disadvantaged children.
Adding in the contribution of parental occupation to these predictions is a little trickier.
Recall that parental occupation is based upon the continuous ISEI index. Also recall that I
define my hypothetical “advantaged” child as having a parent at the (national) 75th percentile
of this continuous index, while a “disadvantaged” child is defined as having highest parental
occupation at the (national) 25 th percentile. Hence to add this into the predicted difference
between “advantaged” and “disadvantaged” groups, one needs to know:
(1) How many ISEI points there are between the 25 th and 75th national percentile
(2) How a one point increase in ISEI changes a child‟s expectations
Note that BOTH of these factors may differ across countries.
If the ISEI index had been entered as a single, simple linear term, one would simply multiply
(1)*(2) above to calculate the contribution parental occupation makes to the difference
between advantaged and disadvantaged groups. However, I have entered the ISEI index as a
series of piece-wise linear components, with a knot at the 50 th percentile. Hence one must
break point (1) and point (2) above into two further components:
45
(1a) How many ISEI points there are between the 25 th and 50th percentile of the
national ISEI distribution
(1b) How many ISEI points there are between the 50 th and 75th percentile of the
national ISEI distribution
(2a) How a one point change in ISEI between the 25 th and 50th national percentile
alters a child‟s expectations
(2b) How a one point change i n ISEI between the 50th and 75th national percentile
alters a child‟s expectations
Now, to calculate the contribution of occupation to the advantaged-disadvantaged expectation
gap, once must sum {(1a*2a) + (1b*2b)}. Information on 1a and 1b (i.e. percentiles of the
ISEI distribution) can be easily calculated from Appendix Table 2 (1a is, for example, simple
the 50th percentile minus the 25th percentile- for England this is 51 minus 35 which equals
16). Details on 2a and 2b are provided in the Appendix Table 5 for the US and Appendix
Table 6 for England (e.g. under the label “occupation spline 26-50th percentile coefficient”
for 2b).
To summerise, to get the difference between the expectations of advantaged and
disadvantaged children for models 1-3, one must sum23:
Parental education “high” coefficient
+
Books in the home “high” coefficient
+
Occupation spline 26-50th percentile coefficient * Number of ISEI points between 25 th and
50th percentile
+
Occupation spline 51-75th percentile coefficient * Number of ISEI points between 50th and
75th percentile
23
The coefficients in the tables below refer to a one point increase in the ISEI index over the relevant range (e.g.
so the coefficient for “Occupation spline 26-50th percentile” refers to how the log-odds of expecting to go to
university change with a one point increase in the ISEI scale that occurs between the 26 th and 50th national
percentile).
46
A worked example for England is given below (and correspond to results presented in section
4 – e.g. Figure 4)
Educational expectations
(no test score control)
Test scores
Coefficient on High (over 100) Books (Ref:
Low)
0.86
40.3
0.70
7.24
0.027*16
1.31*16
0.029*15
1.39*15
2.43
89.3
+
Coefficient on High (tertiary) Parental
Education (Ref: Low)
+
Occupation spline 26-50th percentile
coefficient * Number of ISEI points
between 25th and 50th percentile
+
Occupation spline 51-75th percentile
coefficient * Number of ISEI points
between 51th and 75th percentile
=
Difference (in log odds) between
advantaged and disadvantaged children's
expectations
1 Source: Author‟s calculations using PISA 2003 data. Sample size = 3,817.
47
Appendix Table 1. Distribution of highest parental education across OECD countries
% None
Turkey
Austria
Poland
Northern Ireland
Portugal
Mexico
Switzerland
Ireland
England
Denmark
Italy
New Zealand
Luxembourg
Germany
Norway
Hungary
Slovakia
France
Iceland
Spain
Greece
Czech Republic
Korea
Scotland
Belgium(Flemish)
USA
Canada
Finland
Belgium(French)
Sweden
Australia
Japan
Netherlands
OECD
%
ISCED
1
%
ISCED
2
%
ISCED
3B or
3C
%
ISCED
3A
%
ISCED
4
%
ISCED
5B
%
ISCED
5A +
%
Missing
4
0
1
0
19
8
2
1
1
1
0
3
4
5
0
0
1
2
0
3
0
0
2
3
1
1
0
0
2
2
1
0
31
1
0
1
20
18
2
5
1
0
2
1
9
1
0
0
0
2
2
18
8
0
5
0
2
1
1
3
3
1
1
3
20
5
2
11
16
25
18
10
6
8
22
5
2
8
3
7
2
12
10
7
12
1
14
5
4
4
4
7
4
7
11
3
1
34
21
24
3
3
23
0
22
8
5
13
6
19
4
20
14
22
9
2
4
21
11
17
4
0
0
0
4
7
2
6
23
9
42
4
15
13
7
17
5
12
16
9
8
5
6
16
38
21
11
16
18
37
31
14
15
29
19
21
12
21
16
30
0
6
12
19
0
0
6
27
20
10
19
18
13
15
22
24
17
0
26
10
16
7
0
0
17
16
15
3
10
0
13
0
7
29
7
17
7
14
21
20
18
36
13
20
24
15
36
7
3
11
14
13
13
2
7
23
20
13
21
30
21
21
13
17
13
14
15
17
17
19
19
19
20
20
20
20
20
23
24
24
25
25
26
27
27
28
30
30
30
34
34
36
36
37
39
41
0
2
0
7
2
1
3
2
9
4
1
12
13
10
4
1
1
6
2
5
0
4
1
8
7
3
6
1
8
5
3
0
1
4
10
0
6
27
0
45
7
3
6
10
8
17
11
16
26
4
Notes:
1 Data sorted by the percentage of children who reported either parent as holding an ISCED level 5A+
qualification
2 Figures refer to row percentages.
3 ISCED level 0 refers to no formal school, level 1 is equivalent to primary education only, level 2 is lower
secondary education, level 3B/3C refers to basic vocational education, 3A is upper secondary education, level 4
is post secondary education (either short vocational courses of preparation for tertiary education), level 5B is
specialised vocational education, level 5A is a university education (bachelors degree), while level 6 refers to
doctorates.
48
Appendix Table 2. Distribution of highest parental occupation (ISEI index) across
countries
Percentile
10th
25th
50 th
75th
90th
mean
SD
% Missing
Mexico
Turkey
Portugal
Poland
Spain
Greece
Korea
Austria
Hungary
Ireland
Italy
Luxembourg
Switzerland
Northern Ireland
Denmark
France
Belgium(French)
Germany
Japan
Slovakia
England
Belgium(Flemish)
Canada
Finland
Sweden
Scotland
Czech Republic
Netherlands
New Zealand
Australia
Iceland
USA
24
23
26
23
25
26
29
27
30
29
29
29
29
29
29
29
29
30
33
30
29
29
29
29
30
30
33
30
29
30
29
30
28
29
30
33
30
31
37
34
38
34
34
34
34
34
38
34
37
38
38
37
35
35
38
34
38
40
42
39
40
43
43
40
33
45
39
43
43
46
45
45
45
49
49
50
48
48
51
51
51
51
45
50
51
51
51
51
51
51
51
51
51
52
53
56
54
49
51
53
54
56
51
56
56
57
56
56
55
59
57
59
67
59
55
64
66
66
65
67
66
66
64
67
66
69
67
67
69
66
69
67
70
69
69
69
69
69
70
69
69
69
69
70
70
70
69
69
70
70
69
71
70
70
69
70
69
69
71
71
42
42
43
45
45
46
46
47
48
48
48
48
48
48
49
49
50
50
50
50
50
51
51
51
51
51
52
52
52
53
54
54
19
15
16
15
17
17
13
16
15
16
16
17
16
17
15
17
17
16
15
16
17
17
16
17
16
16
15
16
16
16
17
16
5
12
3
2
4
6
3
4
6
4
2
4
3
6
3
4
6
9
11
4
7
5
7
1
3
4
4
7
16
5
2
6
Norway
34
43
53
69
71
55
15
3
OECD
28
34
49
59
69
48
17
Notes:
1 Data refers to points on the ISEI scale of occupational status, as described in section 3. On this scale, higher
values indicate a more prestigious occupation.
2 Countries sorted by mean ISEI score
5
49
Appendix Table 3. Distribution of the number of books in the home across OECD
countries
% 0-100 books
% Over 100 books
% Missing
Mexico
Turkey
Portugal
Greece
Belgium(Flemish)
Scotland
Northern Ireland
USA
Ireland
France
Poland
Switzerland
Netherlands
Austria
Luxembourg
Italy
Japan
Denmark
Slovakia
England
Belgium(French)
Finland
Germany
Canada
Korea
New Zealand
Spain
Australia
Sweden
Hungary
Norway
Iceland
86
79
68
65
59
58
58
58
58
57
58
56
55
56
54
54
54
52
54
50
51
52
46
43
51
47
47
41
40
41
36
36
10
18
31
34
38
40
40
40
40
41
41
41
42
42
44
45
45
45
45
45
46
47
48
49
49
51
52
56
58
58
61
63
4
3
2
2
4
2
2
2
2
2
1
2
3
2
2
1
1
3
1
5
4
1
6
8
0
3
1
2
2
1
2
2
Czech Republic
33
63
4
55
42
3
OECD
Notes:
1 Data refers to row percentages
2 Data sorted by % over 100 books
50
Appendix Table 4. The proportion of children define as “advantaged” / “disadvantaged”
Main analysis
% Disadvantaged"
Norway
New Zealand
Belgium(French)
Iceland
Netherlands
Australia
England
Belgium(Flemish)
Sweden
Denmark
Finland
Germany
Japan
Czech Republic
Hungary
Scotland
Canada
USA
Spain
Ireland
Luxembourg
Italy
Slovakia
Austria
Northern Ireland
Greece
Korea
Switzerland
France
Mexico
Turkey
Poland
Portugal
3
6
7
7
8
8
9
9
9
9
9
9
10
10
10
11
11
11
11
11
11
12
12
13
14
14
16
16
16
17
19
19
20
% Advantaged
19
16
17
15
14
16
13
12
17
15
17
15
15
19
17
15
13
12
17
13
16
14
13
13
13
13
14
12
12
5
7
12
11
Robustness analysis
% Disadvantaged"
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
% Advantaged
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
51
Appendix Table 5. Parameter estimates US
Model of ..
Educational Expectations
No test score control
With test score control
Beta
SE
Beta
SE
Gender (Ref: Girl)
Boy
Books (Ref: Low)
High
Missing
Parental Education (Ref: Low)
Medium
High
Missing
Occupation spline 0-10th percentile
Occupation spline 11-25th percentile
Occupation spline 26-50th percentile
Occupation spline 50-75th percentile
Occupation spline 76-90th percentile
Occupation spline 91-100th percentile
Immigrant Status (Ref: Native)
Immigrant
Books*Immigrant
High Books, Immigrant
Missing Books, Immigrant
Parental Education*Immigrant
Medium Education, Immigrant
High Education, Immigrant
Missing Education, Immigrant
Occupation 0-10 * Immigrant
Occupation 11-25 * Immigrant
Occupation 26-50 * Immigrant
Occupation 51-75 * Immigrant
Occupation 76-90 * Immigrant
Occupation 91-100 * Immigrant
Constant
Test scores Controlled
Regression technique
Response variable
High Parental Ed
High Books
Occupation 26-50th percentile * 13
Occupation 51-75th percentile * 11
Advantaged - disadvantaged gap
Test Scores
Beta
SE
-0.32
0.07
-0.38
0.07
8.53
2.37
0.58
-0.06
0.08
0.32
0.32
-0.03
0.08
0.31
48.31
-8.72
3.29
14.70
0.13
0.93
0.04
0.009
0.035
0.002
0.005
0.139
-0.015
0.11
0.08
0.29
0.028
0.014
0.012
0.013
0.057
0.016
0.15
0.94
0.21
-0.010
0.028
-0.005
-0.002
0.092
-0.005
0.12
0.09
0.29
0.029
0.014
0.012
0.013
0.058
0.016
-2.97
7.15
-29.86
3.544
1.201
1.303
1.153
8.019
-1.374
4.37
3.45
9.95
1.398
0.598
0.498
0.515
2.015
0.529
-0.43
1.49
-0.74
1.46
65.17
54.19
0.58
-0.06
0.08
0.32
-0.14
-0.80
0.21
0.69
1.13
-8.72
7.04
14.70
0.20
0.17
-0.44
0.03
0.00
0.01
-0.05
0.16
-0.04
-0.66
No
Logistic
Educational
Expectations
0.23
0.21
0.59
0.05
0.03
0.02
0.03
0.13
0.04
0.77
0.24
0.22
0.62
0.05
0.03
0.02
0.03
0.13
0.04
1.16
10.09
1.70
25.13
-2.60
0.06
0.92
-2.42
7.26
-0.07
325.29
NA
OLS
9.57
7.56
21.24
1.93
1.09
0.99
1.29
5.39
1.08
40.55
0.16
0.16
-0.60
0.04
0.00
0.01
-0.04
0.14
-0.04
-2.24
Yes
Logistic
Educational
Expectations
Calculation of the advantaged-disadvantaged gap
0.93
0.94
0.58
0.32
0.02
-0.06
0.06
-0.02
1.59
1.19
PISA Maths test score
7.15
48.31
16.94
12.68
85.08
52
Appendix Table 6. Parameter estimates England
Model of ..
Educational Expectations
No test score control
With test score control
Beta
SE
Beta
SE
Gender (Ref: Girl)
Boy
Books (Ref: Low)
High
Missing
Parental Education (Ref: Low)
Medium
High
Missing
Occupation spline 0-10th percentile
Occupation spline 11-25th percentile
Occupation spline 26-50th percentile
Occupation spline 50-75th percentile
Occupation spline 76-90th percentile
Occupation spline 91-100th percentile
Immigrant Status (Ref: Native)
Immigrant
Books*Immigrant
High Books, Immigrant
Missing Books, Immigrant
Parental Education*Immigrant
Medium Education, Immigrant
High Education, Immigrant
Missing Education, Immigrant
Occupation 0-10 * Immigrant
Occupation 11-25 * Immigrant
Occupation 26-50 * Immigrant
Occupation 51-75 * Immigrant
Occupation 76-90 * Immigrant
Occupation 91-100 * Immigrant
Constant
Test scores Controlled
Regression technique
Response variable
High Parental Ed
High Books
Occupation 26-50th percentile * 16
Occupation 51-75th percentile * 15
Advantaged - disadvantaged gap
Test Scores
Beta
SE
-0.46
0.10
-0.60
0.10
6.10
4.00
0.86
0.35
0.11
0.53
0.49
0.41
0.11
0.50
40.33
-11.96
3.43
18.93
-0.09
0.70
-0.53
0.07
0.00
0.03
0.03
0.04
0.02
0.16
0.12
0.36
0.03
0.05
0.01
0.02
0.05
0.01
0.06
0.71
-0.18
0.08
-0.03
0.01
0.02
0.00
0.02
0.16
0.13
0.36
0.04
0.05
0.01
0.02
0.05
0.02
-11.35
7.24
-34.37
0.65
2.34
1.31
1.39
3.90
-0.26
4.46
4.27
7.55
0.87
1.38
0.39
0.62
1.71
0.69
4.54
2.19
7.24
2.41
-172.45
45.75
-0.46
0.03
0.22
0.91
-0.71
0.58
0.23
1.16
40.33
-11.96
3.43
18.93
0.38
0.28
0.66
0.09
0.13
0.03
0.03
0.10
0.03
3.23
-10.70
0.10
29.71
6.17
2.96
-2.00
-0.98
1.48
0.58
436.30
10.35
8.75
16.37
1.77
3.50
1.01
1.35
3.80
1.29
22.59
-0.01
0.13
0.56
-0.11
0.10
-0.06
-0.03
0.00
-0.02
-4.01
No
Logistic
Educational
Expectations
0.33
0.25
0.62
0.08
0.11
0.03
0.03
0.09
0.03
0.90
0.07
0.21
0.26
-0.20
0.08
-0.04
-0.03
-0.01
-0.03
-14.62
Yes
Logistic
Educational Expectations
Calculation of the advantaged-disadvantaged gap
0.70
0.71
0.86
0.49
0.44
0.22
0.44
0.34
2.43
1.76
NA
OLS
PISA Maths test
score
7.24
40.33
20.91
20.81
89.29